In order to study the regional capital liquidity, Bayesian panel smooth transition regression models were established to address uncertain risk of parameters estimation in PSTR models. Based on the analysis of model statistic structure and the selection of parameters prior, the Metropolis-Hasting within Gibbs sampling method was utilized to estimate model parameters, avoiding the convergent problem when using the nonlinear least square method in PSTR model. The empirical research applies Bayesian PSTR to analyse the panel data of investment and saving in Chinese provinces. The research outcome indicates that the iteration trace of parameters is convergent, and the Metropolis-Hasting within Gibbs sampling method estimates model parameters accurately, showing the effectiveness of Bayesian PSTR model approach.